Gavin J. Bowden

Forecasting chlorine residuals in a water distribution system using a general regression neural network

In a water distribution system (WDS), chlorine disinfection is important in preventing the spread of waterborne diseases. The ability to forecast chlorine residuals at strategic points in the WDS would be a significant aid to water quality managers in helping them to ensure the satisfaction and safety of their customers. In this research, general regression neural networks (GRNNs) are developed for forecasting chlorine residuals in the Myponga WDS, to the south of Adelaide, South Australia, up to 72 h in advance. A number of critical model issues are addressed including: the selection of an appropriate forecasting horizon; the division of the available data into subsets for modelling; and the determination of the inputs relevant to the chlorine forecasts. To determine if the GRNN is able to capture any nonlinear relationships that may be present in the data set, a comparison is made between the GRNN model and a multiple linear regression (MLR) model. Additional investigations are also performed to simulate the effects of a reduced sampling frequency, and to estimate model performance for longer lead-time forecasts. When tested on an independent validation set of data, the GRNN models are able to forecast chlorine levels to a high level of accuracy, up to 72 h in advance. The GRNN also significantly outperforms the MLR model, thereby providing evidence of the existence of nonlinear relationships in the data set.

An Evaluation of Methods for the Selection of Inputs for an Artificial Neural Network Based River Model

Artificial Neural Network (ANN) models are highly flexible function approximators, which have shown their utility in a broad range of ecological modelling applications. The rapid emergence of ANN applications in the field of ecological modelling can be attributed to their advantages over standard statistical approaches. Such flexibility provides a powerful tool for forecasting and prediction, however, the large number of parameters that must be selected only serves to complicate the design process. In most practical circumstances, the design of an ANN is heavily based on heuristic trial-and-error processes with only broad rules of thumb to guide along the way.